Zhiliang comments

Results 14 comments of


                                            Zhiliang

[FR] artifact location when moving mlruns folder

Hi, @dmatrix, if I understand correctly, there is no way to remap the artifact locations when launching the interface with `mlflow ui`. Everything should be predetermined. IMO, this would be...

Distributed GPU training works on A100 but not H100

The problem could be solved by using an AML environment based on `nvidia/cuda:12.8.1-devel-ubuntu24.04`

[BUG] mlflow.log_metric failing for asynchronous logging on distributed training

@daniellok-db @chenmoneygithub Thanks for looking at this! 🙇🏼 @chenmoneygithub The linked code refers to `os.environ['MLFLOW_ASYNC_LOGGING_BUFFERING_SECONDS']`. How does it relate to the batching operations to the metrics? Could you elaborate which...

Unable to specify HYBRID_SHARD for FSDP which requires process group or device_mesh to be passed

> You can use `FSDP2` with `ParallelismConfig` now, such as: > > ``` > fsdp_plugin = FullyShardedDataParallelPlugin(fsdp_version=2,...) > parallelism_config = ParallelismConfig(dp_shard_size=4, dp_replicate_size=2) > ``` > > FSDP1 is not actively...